a p p e n d i x

C
DEMAND FORECASTING
IN MARKETING *

When you finish this appendix you should

• Understand the principles of forecasting.
• Know the differences between Time Series and
Regression Analyses.
• Understand how expert opinion, sales forces and
consumer opinions can all contribute to better
forecasts.

* by Beverley Thompson, The University of Western Sydney, Nepean

 ▼
Demand Forecasting in Mar keting 689

An important part of the marketing planning process is the setting of goals that are
realistic and achievable, given a particular marketing environment and level of marketing
commitment. In marketing, such goals are usually based on market share objectives and
sales targets, both of which require accurate forecasts of total market size, market size of
target segments and likely market share within a targeted segment.

W H AT A R E W E F O R E C A S T I N G ?
Accurate forecasting requires a clear definition of the market in question. Markets may be
differentiated on the basis of the following variables.

GEOGRAPHY
A market may be defined at world, country, state, region, sales territory, town, store or
customer level. When formulating a forecast or other marketing plans, the geographical
dimension must be clearly indicated. Planning Coca Cola consumption for the year 2000
Sydney Olympics for example, will necessitate the forecasting of increased consumption
for the Sydney sales region, but not necessarily for Brisbane.

TIME
A forecast must be defined for a specific time period. Initially several levels of forecast
may be set at differing levels of specificity for the short term, the medium term and the
long term, with increasingly larger confidence intervals around the forecasts, the further
into the future the projection. So for the next year company X may forecast its sales of
product B as being $2.1 million, with a 95 per cent confidence limit of + or −$0.3 million.
However its long range forecast, say for the year 2010 might be $13.5 million, with a
95 per cent confidence interval of + or −$1.8 million. Generally companies will set specific
forecasts on a monthly, quarterly or annual basis.

PRODUCT LEVEL
A forecast may be set for the industry, the company and the product. Even at product
level, separate forecasts might be made for product assortment, product line and product
item. Johnson and Johnson might use survey analysis and demographic data to forecast
the industry sales of the product assortment, shampoo, for the forthcoming year; from this
total industry figure the company might further estimate its market share to create a
company forecast. From this forecast, using past sales projections and company sales
strategy information, Johnson and Johnson might further dissect a forecast for the
product line of children’s shampoo. A further dissection might produce a forecast for
the product item No More Tears in a 500 ml bottle.

AVAILABILITY AND POTENTIAL

In the process of estimating what proportion of the population will ultimately consume
a company’s product, a marketer must estimate the proportion of the population that:

1. wants to buy your assortment, line and item of product and has not
already satisfied their need for this demand in the current time period
2. has the income to pay for your product
3. is interested in buying the brand which your company has for sale.

Thus the estimation of market potential involves consideration of many factors.

 ▼
690 Appendix C

THE DIMENSION OF THE FORECAST

Once a marketer has decided on the geographical, time, product assortment and cus-
tomer availability definitions of the market of interest, the true process of sales forecast-
ing can commence. It is important to estimate market potential before determining
what share a particular company may be able to win with its specific marketing mix.
A common sequence in the forecasting process involves the taking of an increasingly
narrow focus of the market potential by considering the following.

NATIONAL ECONOMIC FORECAST

A marketer should use the work of economists, bankers and government advisers to
gauge the likely economic environment of the period for which the forecast is being pre-
pared, for the country to which the forecast applies.

INDUSTRY FORECAST
An industry forecast considers not only the overall state of the economy but also factors
specific to the industry in which a company is operating. In the case of CSR Building
Materials, management will not only be interested in the overall state of the economy, but
also in the state of the building sector specifically. Some economic effects might be lagged,
for example, so that their positive or negative effects on a particular segment might be
considerably delayed. This is often the case with changes in interest rates, which might
affect the number of new housing starts but not building in progress. In developing an
industry forecast, consideration must also be given to the level of competition, both
within the particular industry and confronting the industry from the outside in the form
of substitute products or foreign imports. Whilst a marketer generally does not have to
develop forecasts for an entire industry, the onus is on he or she to have a clear under-
standing of the parameters of the industry in which he or she is operating.

COMPANY OR PRODUCT FORECAST

SALES POTENTIAL A specific company or product forecast ultimately involves a consideration of the sales
The maximum share of a potential—the maximum share of a particular market that a product can expect to gain
particular market that a in a given period. This obviously hinges on the overall market potential and, within the
product can expect to gain market, on the company’s marketing effort and potential. For example, when Lyric Opera
in a given period in Brisbane launched its ‘Opera for the under 35s’ campaign, its promotion boosted the
sales potential not only for its own cultural product but also for other cultural products
in the broader arts market.
Sales potential may be estimated using either the break-down approach or the
build-up approach. The break-down approach considers a downward-expanding pyramid
structure breakdown of economy-market-company-product forecast, thus finishing with
the smallest unit. The build-up approach begins by estimating the consumption of a spe-
cific user or buyer in a particular geographic area, in a given time period. This amount is
then progressively cumulated over all potential buyers in the area, and then progressively
for the entire sales territory, region or country.
A specific product forecast must also consider the product in the context of its stage
in the product lifecycle. A product at the introduction stage will expect to have different
rates of growth and volume than will a product at the maturity stage. In the case of new
products or new technologies, the forecast must also consider the likely rate of change
of acceptance of the new product. Market research will be a necessary input into the
forecasting process for new products, and indeed for many pre-existing products.
 ▼
Demand Forecasting in Mar keting 691

SALES FORECASTING TECHNIQUES

A number of forecasting methods is available to the marketer, some of which are scien-
tific and some more judgmental. In practice a combination of methods may well be used,
even for the one product. The choice of method will be influenced by the nature of the
product and its market, the stage in the product lifecycle and the nature of the product’s
historical sales trend, the time and budget available for the forecasting procedure, the
purpose and importance of the forecast, the availability of background information, the
expertise and experience of the forecaster, and the stability or otherwise of conditions in
the projected marketing environment.

TIME SERIES ANALYSIS

Time-series analysis is mentioned first because the prerequisite collection of sales data,
graphed against time, is useful in its own right as an analytical tool for ascertaining cur-
rent market demand, as well as providing the base for time-series forecasting.
In all forms of statistical analysis, ‘a picture tells a thousand words’. Much feel for the
market situation of a product can be gleaned from a graph. In this age of personal com-
puters, with a wealth of spreadsheet, statistical and graphical packages, no marketing per-
son should operate without the assistance of a graphical depiction of sales trends (using
line graphs or histograms), market share (using pie charts or bar graphs) or statistical
comparisons (using such tools as stem-and-leaf plots).

Examples of graphical analysis

Line chart Histogram Pie chart

* * *
Stem-and-Leaf Plot

Forecasting based on time-series uses the company’s historical sales data to search for
certain patterns which are projected forward in time to produce a future forecast. There
are four components in this type of analysis:

1. Trend describes the underlying growth or decline pattern in the sales. This is
the major thrust of sales progress. It may also reflect strong underlying factors,
such as economic, demographic or technological trends. A linear trend is the
simplest trend to understand mathematically. However, non-linear forms such
as quadratic or exponential trends may also be estimated using mathematical
models and time-series data.
 ▼
692 Appendix C

Examples of trends
(a) Linear—for example, a product experiencing a consistent rate of growth, or decline.

(b) Positive quadratic—for example, a product with a revamped marketing mix.

(c) Negative quadratic—for example, a product with a compact life cycle.

(d) Exponential growth curve—for example, a product enjoying an increasing

rate of growth.

(e) Gompertz curve—for example, a product experiencing a levelling off in growth.

This is very common for many mature products in the marketplace.
 ▼
Demand Forecasting in Mar keting 693

2. Season refers to movements in a series which occur at regular times over the
years. This pattern of movement in the data may be detected in quarterly,
monthly, daily or even hourly data and may be related to the weather or
climatic seasons, to patterns of business activity (such as a flurry of activity at
the end of June because of the end of the financial year) or to patterns of
consumer behaviour (such as increased consumption of toys at Christmas).

3. Cycle captures the longer swings up and down which are often caused by
economic upswings and downturns, and similar cyclical swings in consumer
confidence, consumer tastes, fads and competitive effects. Cyclical effects are
difficult to predict because they do not recur on a regular basis. However,
approximate cyclical effects can be built into medium- and long-range forecasts.

4. Irregular component analysis is an attempt to allow for random, erratic or

unexpected events such as political unrest or wars, natural disasters or financial
bombshells. Since these effects are unusual and unpredictable, they are usually
removed from past data before conducting any other analyses.

Time-series analysis involves defining the data series components of trend, cycle, TIME-SERIES ANALYSIS
season and irregularity and combining them to form a sales forecast. This technique is Defining the data series
particularly effective for forecasting sales of products with relatively stable demand. Time components of trend, cycle,
series analysis treats sales as being a function of time which in fact acts as a proxy for a season and irregularity and
multitude of other variables, such as competition, consumer preferences or economic combining them to form
activity. a sales forecast
Sometimes, however, we may prefer to deal with these variables directly in making
our forecast. Relationship methods enable us to do just this.

RELATIONSHIP METHODS
Leading indicators The demand for many products is derived from some other activity, LEADING INDICATORS
the statistics for which can provide the basis of a forecast for the product in question. For Statistics which can provide
example, CSR Building Materials has much of its demand derived from housing starts and the basis of a forecast for the
building approvals, which in turn are dependent to a large extent on the health of the product in question
economy. Many leading indicators are supplied to the building industry through the
Housing Industry Association, the Indicative Planning Association and BIS Shrapnel.
Economic indicators such as the Consumer Price Index (CPI) and the All-ordinaries Share
Index are also available for use in forecasting.
Regression analysis This enables us to express sales as a variable which is affected REGRESSION ANALYSIS
by (or dependent on) one or more other (independent) variables. We might conclude, for Analysis which allows for the
example, that demand for CSR bricks will be at least partly dependent on the number of expression of sales as a variable
housing starts in some preceding periods. which is affected by one or
Although we mentioned numerous types of trends in the previous section, we will more variables
concentrate here only on the linear trend.
 ▼
694 Appendix C

Notation:
The dependent variable, e.g. sales, is usually denoted by Y
The independent variables are denoted by X1, X2...Xn

Y * denotes the ‘fitted’ Y values, when the original X values are substituted into the
regression equation
In statistical terminology we say that sales is a function of one or more independent
variables:
Y = f( X1, X2...Xn )

SIMPLE REGRESSION MODEL
A regression model with only
one independent variable
Simple linear regression refers to a regression model with only one independent
variable. Multiple regression models have more than one independent variable.
We will discuss here only the simple regression model.
The form of the simple linear equation is:
Y * = a + b 1X 1
where:a = the intercept term (the point at which the line crosses the Y axis or
where X = 0)
b = the slope coefficient. As X increases 1 unit, Y * increases by b units.
If b is negative, the equation will be that of a negatively-sloped line—that is,
as X increases, Y * will decrease.

Figure 1 Y
A theoretical regression
line showing the role of Y * = a + bX
the slope and intercept
b
a

X=0 1 unit X

Formulae for estimating the simple linear regression equation are as follows:

b = Σ (Xt – X) (Yt – Y) a=Y–bX

Σ (Xt – X)2

Example: Assume that the sales of a particular type of CSR roof tile is dependent on
the number of housing starts in the preceding half year. Other factors of potential
importance will be ignored in this simple equation. The following (fictitious) data shown
in Figure 2 might be considered, and are represented graphically in Figure 3.

Figure 2
Hypothetical data for PERIOD X = HOUSING STARTS IN Y = SALES OF CSR
CSR roof tile sales PRECEDING HALF YEAR ROOF TILES
(THOUSANDS) (MILLIONS)
1 100 43
2 112 54
3 165 91
4 150 80
5 121 60
 ▼
Demand Forecasting in Mar keting 695

Y Figure 3

Sales of roof tiles

100 Graphical representation
x
(millions)
of data
80 x

60 x
x
40 x

X
100 120 140 150 160

Housing starts ’000

The regression involves a number of intermediate calculations, the detail of which is
shown in Figure 4.

Figure 4
PERIOD X X – X (X – X)2 Y Y – Y (X–X) (Y–Y) Intermediate regression
calculation
1 100 –29.6 876.16 43 –22.6 668.96
2 112 –17.6 309.76 54 –11.6 204.16
3 165 35.4 253.16 91 25.4 899.16
4 150 20.4 416.16 80 14.4 293.76
5 121 –8.6 73.96 60 –5.6 48.16

ΣX = 648 Σ(X–X)2 ΣY = 328 Σ(X–X)(Y–Y)

X = Σ X/n = 2929.2 Y = ΣY/n = 2114.2
= 648/5 = 328/5
= 129.6 = 65.6

b = Σ (Xt – X) (Yt – Y) = 2114.2 = 0.7218

Σ(X t –X)2 2929.2

a = Y – bX = 65.6 – (0.7218)129.6
= 65.6 – 93.55
= –27.95

This provides us with the final equation:

Y * = –27.95 + 0.7218 X
which can be interpreted as:
b = 0.7218

This means that for every 1 unit of X (every 1000 housing starts), sales of roof tiles will
= 0.7218 units of Y (0.7218 x 1 million roof tiles = 721 800 roof tiles).

a = –27.95 has a less significant meaning, being negative. This sometimes happens in
a mathematical regression equation, with a positively-sloped line.
This is an example based on fictitious data, but it serves to show the workings of
a regression equation. In the real world of business, the tedium of such calculation
is removed by the use of statistical and forecasting computer software, such as SPSS,
Microtab, TSP and Forecast-Pro.
 ▼
696 Appendix C

Many more statistics can be calculated in relation to regression analysis, which enables
hypotheses to be tested relating to the importance of variables used in the equation and
to the overall value of the equation as a whole. However, as with any complex statistical
methodology, caution must be used in the design, conduct and analysis, as incorrect initial
assumptions can invalidate the entire process. Nonetheless, in the hands of experienced
and knowledgeable users, and with the use of increasingly available and sophisticated
computer technology, regression techniques provide a useful quantitative basis for many
business forecasts. Further details of regression analysis can be obtained from statistics or
econometric textbooks.

COMBINATION OF TIME SERIES

AND REGRESSION ANALYSIS
As mentioned above, simple regression involves a regression model with only one depen-
dent variable. However, this is often too simplistic to tell the entire story. Astute use of
multiple regression requires much skill in selecting appropriate variables and in main-
taining the statistical validity of the more complex relationship. Sometimes forecasters
substitute time for multiple independent variables in a regression equation as a proxy for
a multitude of other variables.
The following example, using quarterly champagne sales (fictitious data), shows how
time series analysis and regression analysis can be used to create a forecast, with the addi-
tion of seasonal indices. The steps are as follows:

1. Compilation of the appropriate background data. This method assumes that at

least ten periods of data are available. In this example we have chosen to use
quarterly data. This data is shown as the sales column in Figure 5.

2. Smoothing of data to form an approximate trend, from which the seasonal

indices can eventually be defined. This procedure commences with the calcula-
tion of a four-quarter moving total, then the calculation of a centred moving
total, which is then divided by four to give a centred moving average. The four
period moving average can be used as a trend in its own right. These calculations
are shown in Figure 5 (four-quarter MT, centred MT and centred MA columns).

3. The final column of Figure 5 shows the index (sales/CMA). Values below 100
show that sales are less than the centred moving average, and vice versa. Figure 6
summarises the index column from Figure 5, and calculates the average index for
each of the respective quarters. A final adjustment ensures that all four indices
add up to 100. A quick check of the dimensions of these seasonal indices con-
firms the expected seasonal pattern of champagne sales, with the highest index
value occurring in quarter IV—the festive season.

4. Calculation of regression equation: sales = function (time). The result is shown

in Figure 7.

5. Application of seasonal indices to projected trend line to produce forecast for

next eight quarters. Figure 8 shows the workings of this, while Figure 9 provides
a graphical representation of the sales and forecast.
 ▼
Demand Forecasting in Mar keting 697

Figure 5
KEY MT = moving total
Fictional data of sales of a
MA = moving average champagne product in A$
CMA = centred moving average million.
SI = seasonal index
T = trend

PERIOD SALES FOUR-QUARTER CENTRED CENTRED INDEX

MT MT MA SALES/CMA
Year 1 Quarter I 0.8
II 1.3
III 1.4 5.8 6.1 1.525 91.8
IV 2.3 6.4 6.5 1.625 141.54

Year 2 Quarter I 1.4 6.6 6.9 1.725 81.16

II 1.5 7.2 7.6 1.9 78.95
III 2.0 8.0 8.0 4.0 50
IV 3.1 8.0 8.15 2.04 151.96

Year 3 Quarter I 1.4 8.3 8.45 2.11 66.35

II 1.8 8.6 8.85 2.21 81.44
III 2.3 9.1 9.4 2.35 97.87
IV 3.6 9.7 10.35 2.59 139.0

Year 4 Quarter I 2.0 11.0 11.75 2.94 68.03

II 3.1 12.5 12.85 3.21 96.57
III 3.8 13.2
IV 4.3

Figure 6
Quarter I II III IV Calculation of seasonal
index (using index column
91.80 141.54 from Figure 5)
81.16 78.95 50.00 151.96
66.35 81.44 97.87 139.00
68.03 96.57

Average 71.85 85.65 79.89 144.16

Adjusted
Final SI 75.33 89.79 83.75 151.13 Total = 400

Figure 7
Calculation of regression trend
Calculation of regression
trend for champagne
Let Y = Sales data as above sales forecast
Let X = Time from 1 to 16

Using a regression equation fitted to the above data:

Y * = 0.7787 + 0.1738 t

The interpretation of the b value of 0.1738 is that for every increase in time period, sales
will increase by 0.1738 million dollars or $173,800.
 ▼
698 Appendix C

Let us now assume that we wish to forecast two years into the future beyond the
given data. Given that our data stops at period 16, we will wish to be forecasting sales for
periods 17 to 24. Thus in the equation Y* = 0.7787 + 0.1738t, we will be substituting pro-
gressively the values t = 17, t = 18...t = 24 to get the trend base of the forecast. We will
then multiply this by the seasonal index to obtain the final forecast.

Figure 8
Combining regression PERIOD QUARTER TREND SEASONAL FINAL FORECAST
trend and seasonal indices NUMBER VALUE INDEX/100 TXSI
to create final forecast
17 I 3.73 0.7533 2.81
18 II 3.91 0.8979 3.51
19 III 4.08 0.8375 3.42
20 IV 4.25 1.5113 6.42
21 I 4.43 0.7533 3.34
22 II 4.60 0.8979 4.13
23 III 4.78 0.8375 4.00
24 IV 4.95 1.5113 6.80

It is important to note that this method presupposes that a sufficiently long sales his-
tory exists to be able to develop a seasonal index, and that pre-existing conditions will
hold for the next eight periods. In reality this may not be so, but this statistical forecast
base can then be adjusted for the findings of qualitative research. The overlay of a cycli-
cal effect may also be necessary. However, despite the fact that such techniques might
have some shortcomings, they are usually much more satisfactory than pure ‘guessti-
mates’. In practice, adjustments to forecasts obtained through quantitative methods may
be made on the basis of more qualitative methods. Some of these will now be discussed.

EXPERT OPINION
At times, forecasts may be made with little formal statistical analysis, relying instead on
the use of experienced judgment by executives, marketing consultants, trade associations,
buyers/sellers or academics.
Forecasts made by executives may be based on intuition rather than scientific
methodology. Their success probably depends on the degree of insight into the overall
market behaviour of the particular executive. In a large organisation, reliance on such
judgment without the backup of more rigorous methods may prove to be risky in the
long term, especially in a fluid competitive environment.

Figure 9
Sales in $million

Graphic representation of
original sales, and forecast 6
Actual
➤
➤

4
➤
➤
2 Forecast

Year 1 Year 2 Year 3 Year 4 Year 5 Year 6

Forecasts made by in-market contacts, such as customers, distributors, dealers or

wholesalers, may also suffer from anticipation bias, whereby sales estimates are inflated
in anticipation of unrealistically high future business.
 ▼
Demand Forecasting in Mar keting 699

Independent expert opinions from marketing consultants, academics or trade associa-

tions are also worthwhile obtaining because of their specialist skills and database access.
However, even experts can be wrong, especially if they are not in close contact with
‘the real world’. One precautionary approach used, especially for particularly critical
forecasts, is a jury of expert opinion, whereby a consensus of opinion is reached from JURY OF EXPERT OPINION
the opinions of several experts, each with different spheres of expertise or experience. When consensus of opinion is
For example, the views of the marketing executives, economists, software research and reached from the opinions of
development experts and Bill Gates himself may be merged to help to ascertain the several experts
forecast demand for Microsoft products in forthcoming periods. One such approach,
coined the Delphi technique, involves gathering managers’ and others’ opinions THE DELPHI TECHNIQUE
independently, validating them centrally, and progressively refining them in successive The independent gathering
rounds until consensus is reached. The Delphi method avoids many weighting and of managers’ and others
judgmental problems, with the median of the groups’ overall responses forming the final opinions, their central
forecast. The approach is particularly useful for new product forecasts when no historical validation and reinforcement
information is available on which to base a forecast. Overall, however, expert opinion in successive rounds until
should not be used in isolation, but as a complement to quantitative techniques. consensus is reached

SALES FORCE OPINIONS AND

CUSTOMER OPINIONS
Two other sources of market opinion are often sought in the forecasting process. These
are sales force opinions and actual/potential customer opinions.
The opinions of salespeople should be considered to some extent, as they are at the
forefront of the market and in an ideal position to assess customer reactions and com-
petitors’ activities. Their estimates are especially useful in some business markets where
the customers may be well-known to the salespeople and where relatively few customers
dominate the market. Good retail sales assistants opinions should also not be ignored.
However, salespeoples’ estimates should be used with caution. They are often based on
a relatively narrow perspective of the entire market, with little consideration of broader
economic trends or even of wider policy effects from within the organisation. Certain
biases may also be present in salespeoples’ estimates, either because of their personal opti-
mism or pessimism or because of their short-term emphasis on recent successes or failures
in the marketplace. There is also a risk that that they may underestimate demand in order
to be allocated low sales quotas, which form the benchmark for the payment of bonuses.
Salesperson representation in the jury of expert opinion provides a two-way benefit,
not only through receipt of their opinion, but also in their appreciation of having their
opinion valued. This, combined with the payment of salaries to salespeople with reduced
emphasis on bonuses, acts as an incentive for salespeople to provide accurate estimates.
Actual or potential customer opinion can be ascertained via the market research process.
Surveys and panel analysis of opinions of past, current and potential customers, along
with surveys or focus group research of retailers, wholesalers, agents and other gatekeep-
ers, can help to identify what is happening in different market segments. This sort of
activity is particularly necessary in a dynamic market.
Surveys are sometimes combined with market tests when the company wants to esti-
mate customers’ reactions to possible changes in its marketing mix. Pricing experiments,
new product testing and even advertisement testing can often be undertaken in an
economical way in a small test market. Such tests may simply involve a measurement of
likely demand for a product the marketing mix of which is almost finalised. Alternatively,
it may involve some actual experimentation with pricing or promotion strategies.